复杂场景下基于改进的Y0L0v5-pose的异常行为检测研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）07-0071-06

Abstract： This paper proposes an abnormal behavior detection algorithm based on the improved YOLOv5-pose in complexscenes.ItusesFPTtoreplacetheFPN+PANmodule，enablingthefeaturemapstoachieve globalandlocalinteraction acrossscales and spaces，and improving the accuracyofjoint point detection.Inthe Neck module，askipconnection structure is employedto efectively fusethe information ofthe input featuresand the multi-scale features output through the network， improvingtheabilitytocapturedetailed informationandenancing theauracyofdetectingoluded jointpoints.Experimental results show that the improved algorithm achieves an average accuracy of 9 9 . 5 % on the CrowdPose dataset，which is 2 . 4 % higher thanthatoftheoriginalmodel.Theimprovedmodelnotonlyhashigherdetectionaccuracybutalsosignificantlyimprovesthe recognition performance of small targets.

Keywords： YOLOv5-pose; behavior recognition; joint point detection; FPT; skip connection

0 引言

在城市人员复杂、社会安全风险高的重点场所，极易发生群体冲突、极端暴力等社会安全突发事件，加剧民众的心理恐慌，严重影响到社会稳定。（剩余8613字）